1C Litecode MCP

A local knowledge graph of a 1C configuration for an agent: Memgraph and ONNX embeddings with no cloud, for bases with hundreds of thousands of routines

MCP server

Low risk

We rate an entry low when it mostly gives the agent instructions and reference material.

Why this level

  • Only works with the configuration report export and source files, does not connect to a live base
  • All computation and storage are local, code and metadata are sent nowhere
All reasons and checks
Russian stack

svhov/1c-litecode-mcp

Install

Manual install

docker compose build && docker compose up -d memgraph && docker compose up -d litecode-my_project

Run inside lite/ after preparing docker-compose.yml from the example with paths to your own data.

This is third-party code. Review the repository files before installing.

What it does

The server parses a 1C configuration report, BSL source files, forms, roles and a GUID mapping, loads everything into a Memgraph graph (metadata objects, attributes, procedure calls, USED_IN and MOVEMENTS_IN links) and indexes it semantically with local ONNX E5-base and cross-encoder models with no calls to external APIs. An agent gets two tools: search_metadata for structural graph search (14 operations: browsing, an object card, a call tree, role rights, HTTP services and more) and search_by_embedding for meaning-based search when the exact object name is unknown. It runs in a single Docker container on Memgraph plus sqlite-vec, uses less memory than Neo4j with PyTorch, and can index a configuration with 638,000 BSL routines in streaming mode without running out of memory.

Who it is for. For 1C developers with large configurations who need an agent to quickly find the right object, procedure or relationship without manually digging through the configuration report.

Good fit when

  • The configuration is large (thousands of objects, tens of thousands of routines) and an agent gets lost without structure
  • It matters that code and metadata never leave the local machine or network
  • You need both exact structural search (who calls this routine) and meaning-based search

Not a fit when

  • You have no configuration report and no exported source files to parse: the server builds the graph from those
  • You cannot allocate 2.5 GB of RAM for peak indexing on large bases
  • You need exact case handling for Cyrillic text in graph queries: Memgraph has a known toLower() limitation there

Example request

Find, by meaning, where payroll postings are generated in the configuration and show that routine's body

Limitations

The ONNX models for semantic search (E5-base, cross-encoder) are not bundled and are downloaded separately; without them ENABLE_EMBEDDING=true fails at startup. On CPU without a GPU, indexing 600,000 routines takes 10-18 hours, though it resumes safely after a restart. Memgraph does not handle Cyrillic correctly in toLower(), and search compensates with multiple case variants on the Python side.

How to disable. Stop the containers with docker compose down and remove the project's server from your MCP client config.

MCP

Transport
sse
Authentication
not required
Environment variables
Environment variables
PROJECT_NAME
required
Unique project identifier for isolating data in the shared Memgraph.
ENABLE_EMBEDDING
Turns on semantic search, requires the ONNX models to be downloaded in advance.
MEMGRAPH_URI
Bolt URI for connecting to Memgraph, defaults to bolt://memgraph:7687.

Security check

  • Only works with the configuration report export and source files, does not connect to a live base
  • All computation and storage are local, code and metadata are sent nowhere

README in short

The README compares Lite in detail against a Neo4j and PyTorch setup by memory, speed and stability, gives a 60-second demo launch, describes the architecture (Memgraph, FastMCP over SSE, E5-base, cross-encoder, sqlite-vec), the full list of MCP operations for structural and semantic search, a built-in benchmark pipeline with P@1/P@3/P@5/MRR metrics, an environment variable table, the graph model, and known limitations around Cyrillic handling and CPU indexing speed.

FAQ

Does it need access to a live 1C base?

No, the server works with the configuration report and exported source files, and does not connect to a working base.

Can several projects run at once?

Yes, all projects share one Memgraph and are isolated by PROJECT_NAME, each with its own MCP service and port.

Editors’ pick

A skills library that gives coding agents a development process: brainstorming, planning, TDD, subagents and code review

PluginMedium riskNo VPN needed292.5KRepository stars
Editors’ pick

Small composable skills for engineering with agents: plan grilling, TDD, bug diagnosis, code review and architecture

SkillLow risk271.4KRepository stars
Editors’ pick

GitHub toolkit for spec-driven development: the specify CLI adds agent commands and skills to a project, from principles to implementation

CLIMedium riskNo VPN needed139.3KRepository stars

Reference MCP servers

Model Context Protocol servers

Official

Official reference MCP servers: Filesystem, Fetch, Git, Memory, Sequential Thinking, Time and Everything

MCP serverMedium risk90.6KRepository stars
Foxx AI1C Litecode MCP

I am Foxx AI and I have already vetted this tool. Ask about install, setup or anything else, and I will keep it simple.